Vis & Society Assignment 3

Proposition: Secondary education teachers are the key to higher incomes for a country.

FOR the Proposition

Number of secondary teachers compared to GNI

Design Decisions and Rationale:

(2) There is a high correlation between these two datasets and it is easy to see

Finding a good part of the data is important and useful. I used a correlations between every pair of columns to find the most correlating data between education and economic data, and this is what I found.

(2) There are not any visible outliers

Again, finding a nice clean column of data is useful. The audience can trust the data more because of the strong correlation, and I don't need to account for any weird outliers.

(-2) The correlating points are easy to see, and the countries with low incomes and smaller countries are ignored because their points are all squished at the origin

This is the part that makes the visualization very deceptive, because it ignores a huge amount of the data, while still technically displaying it. This emphasizes higher income and larger countries, and almost completely ignores some of them. It's hard to see that they're being ignored, though, because they are still there.

AGAINST the Proposition

Trained teachers in secondary education compared to GNI

Design Decisions and Rationale:

(2) The lack of correlation is very clear visually in the graph

This graph only needs to disprove the claim, so showing no correlation does a good job at this. It is clear that there is no correlation because you can see many points going in different directions.

(1) It expands the points at very low populations, unlike the prior graph, because it normalizes the numbers of teachers

Accounting for population by using percentage of teachers spreads out many of the low GNI points. It probably would be good to change to a log scale as well on GNI to really show it, but thsi shows no correlation enough.

(1) The columns are not adjusted, so it seems trustworthy

Adjusting columns can seem very misleading, because the default assumption is that things are linear. However, in some cases you can hide data with an unadjusted column if all the points are clustered on one end.

Final Reflection

I spent most of the time pouring over the different datasets, trying to find some good correlation. I ended up merging education and economics data and using pandas df.corr() method to find any correlating columns. That way I could end up with a scatterplot with nice correlations. It was interesting that I didn't mean to make a misleading plot at first, but ended up with one because the correlation is less affected by the cluster of points with lower GNI and lower population. I think it's easy to trust the numbers and not realize that you need to think about what makes sense, and how the data is assembled. Using raw population, for example, is not great because it doesn't account for country size. Using GNI is probably good, but I'm not an expert on how they calculate it, so their might be issues with it. Sometimes it feels like you are better off using simpler calculations that are more transparent, such as percentage of population, instead of highly processed values like GNI.

I didn't need to do that much with the visualization to make it deceptive. I also didn't really want to, and it would make it seem a lot less trustworthy if the visualization had a bunch of weird suspicious things like adjusted axes. I think having a simpler graph makes it more likely that you can deceive someone because you can hide under the cover of seeming transparency.